An Effective Investment Strategy Using Deep Graph Reinforcement Learning : Evidence from China's Stock Market

Deep learning model has great potential and advantages in modeling the relations between stocks and helping trading on financial market. This paper studies the application of deep learning, reinforcement learning and graph neural network methods in portfolio optimization. A graph reinforcement learning model is proposed for portfolio construction, which combines graph neural network with deep reinforcement learning. With the return of portfolio taken as the optimization goal, both the historical data and interaction of stock are considered to determine the optimal investment weight. The experiment in China's A-share stock market compares the returns of portfolio formed by our model with benchmark index and those formed by other deep learning models. The results show that with the increase of transaction costs, the performance of the investment strategy based on graph reinforcement learning is not affected and has a certain robustness. Different from previous model, deep graph reinforcement learning (DGRL) model can make profit as much as possible and pay attention to risk control at the same time, so that the trading profit rollback is smaller and the model performance is more stable